SLIDE 1
Algorithmic Bias in Google Searches for Political Parties and Candidates
Johannes Nakayama, Nils Plettenberg, Laura Burbach, André Calero Valdez Human-Computer Interaction Center, RWTH Aachen University
Datenspende BTW17
Algorithmic Bias in Google Searches for Political Parties and - - PowerPoint PPT Presentation
Algorithmic Bias in Google Searches for Political Parties and - - PowerPoint PPT Presentation
Algorithmic Bias in Google Searches for Political Parties and Candidates Johannes Nakayama, Nils Plettenberg, Laura Burbach, Andr Calero Valdez Human-Computer Interaction Center, RWTH Aachen University The Datenspende BTW17 Dataset search
SLIDE 2
SLIDE 3
database search terms
Ø 1647 searches per day
…
search terms search terms 4000 participants
SLIDE 4
parties:
‘CDU’ ‘CSU’ ‘SPD’ ‘FDP’ ‘Bündnis 90/Die Grünen’ ‘Die Linke‘ ‘AfD’
candidates:
‘Angela Merkel’ ‘Martin Schulz’ ‘Christian Lindner’ ‘Katrin Göring-Eckardt’ ‘Cem Özdemir’ ‘Sahra Wagenknecht’ ‘Dietmar Bartsch’ ‘Alice Weidel’ ‘Alexander Gauland’
SLIDE 5
Can we find evidence for personalization in the dataset?
SLIDE 6
First approach
set intersection: first three/six/nine results
SLIDE 7
C D A G F B H E I A B C D E F G H I
SLIDE 8
C D A G F B H E I A B C D E F G H I
SLIDE 9
- verlap of 66.67 %
repeat for every list with every other list compute mean
C D A A B C
SLIDE 10
SLIDE 11
downside: only rough order effects
SLIDE 12
SLIDE 13
Rank-biased overlap (RBO)
Result 1 Result 2 Result 3 Result 4 Result 5 Result 6 Result 7 Result 8
Result 1
Result 2
Result 3
Result 4
Result 5
Result 6
Result 7
Result 8 (Webber et al. 2010)
SLIDE 14
How does RBO work?
User 1 User 2 A A B B C C D D E E F F G G H H I I J J
RBO score: 1.0
User 1 User 2 A K B L C M D N E O F P G Q H R I S J T
RBO score: 0
User 1 User 2 A J B I C H D G E F F E G D H C I B J A
Identical Lists Entirely different items Same items, reverse
- rder
RBO score: 0.5116076
SLIDE 15
User 1 User 2 A K B B C C D D E E F F G G H H I I J J
How does RBO work?
RBO score: 0.7297158
User 1 User 2 A A B B C K D D E E F F G G H H I I J J
RBO score: 0.901564
User 1 User 2 A A B B C C D D E E F F G G H H I I J J K P L Q M R N S O T
1st item different 3rd item different Last 5 items different RBO score: 0.8747158
SLIDE 16
Similarity of Search Results
SLIDE 17
SLIDE 18
SLIDE 19
SLIDE 20
SLIDE 21
T-test
significant differences between RBO scores of parties and candidates t = 27.4, p < .001
SLIDE 22
SLIDE 23
H
- w
c a n w e e x p l a i n t h e d
- w
n w a r d t r e n d ?
SLIDE 24
SLIDE 25
Search Volume
Trends
no absolute values
- nly relative scores
SLIDE 26
SLIDE 27
SLIDE 28
Summary
significant difference between parties and candidates dependence between RBO scores and search volume
SLIDE 29
Future Work
RBO package (already in progress)
R
content analysis
T
location
SLIDE 30
- pen data and all analyses available at:
- sf.io/e598k